A general algorithm for approximate inference in multiply sectioned bayesian networks

Abstract

Multiply Sectioned Bayesian Networks(MSBNs) extend the junction tree based inference algorithms into a coherent framework for flexible modelling and effective inference in large domains. However, these junction tree based algorithms are limited by the need to maintain an exact representation of clique potentials. This paper presents a new unified inference framework for MSBNs that combines approximate inference algorithms and junction tree based inference algorithms, thereby circumvents this limitation. As a result our algorithm allow inference in much larger domains given the same computational resources. We believe it is the very first approximate inference algorithm for MSBNs.